[c-nsp] what is it with 3550s?

[Jon Lewis]

On Sun, 21 Feb 2010, TCIS List Acct wrote:

> We've got a boatload of 3550-EMI's (for colo/server aggregation duties) and 
> are looking at replacing them in the next 12-24 months  The C3750G-24/48-E 
> series seem to be a good upgrade path (all gig ports, layer3 routing, IPv6 
> support, fairly easy to source on the used market) -- curious as to why you 
> said they didn't look viable...

I haven't played with the 3750 line, but assuming it's similar in software 
to the 3560, the trouble with them (vs the 3550), particularly in a 
colo/server aggregation setup, is a serious lack of flexibility in 
per-port policing.

With the 3550, you can use input/output service-policy to police each port 
to arbitrary bandwidh.  If you want to limit a customer to 2mbit/s in both 
directions, it's trivial to do.

With the newer switches, cisco no longer supports per-port egress 
policing.  Instead, you have srr-queue bandwidth limit [10-90], which 
limits the egress port speed to 10-90% of physical speed.  i.e. srr-queue 
bandwidth 10 on a 100mbit port results in 10mbit/s.  That would be the 
lowest "policed" rate you can configure for egress.  If you can live with 
that, then it's not so bad, because you can come up with the %'s for all 
the other rates you'd want to limit ports to.

Also, because the egress bandwidth limit is configured for the "whole 
port" you can't use service-policy and ACLs to police some traffic but not 
other, or police certain types of traffic to different rates on the same 
port.

----------------------------------------------------------------------
  Jon Lewis                   |  I route
  Senior Network Engineer     |  therefore you are
  Atlantic Net                |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________